View
213
Download
0
Embed Size (px)
Citation preview
1
Mobility Management in
Sensor Networks
Muneeb Ali†‡, Thiemo Voigt‡, and Zartash Uzmi†
†LUMS, Pakistan‡SICS, Sweden
2
Two recent research trends that motivate our work:
(a) Towards a Sensor Network architecture
(b) Mobility in Sensor Networks
● Towards a Sensor Network Architecture - Internet vs Sensor Networks
- Sensor-Net Protocol (SP)
● Mobility in Sensor Networks
● Mobility-Management in Sensor Networks
● On-going Work
● Open Issues
● Conclusion
Outline
3
Internet vs Sensor-Nets
The Internet
● Independent hosts● End to end flows● Two tier architecture● Wired (generally)● Latency● Throughput● Bandwidth is relatively cheap
Sensor Networks
● Collaborative use
● Collect, disseminate, ...
● Ad-hoc (more homogeneous)
● Low power wireless
● Wake time
● Very low utilization
● Bandwidth is expensive
Reference: Philip Lewis, ICSI Talk, May 2004
4
Internet vs Sensor-Nets
Lessons Learned
● Internet solutions generally do not apply to sensor networks
● Their underlying techniques do
● Apply, change and adapt to the peculiarities of sensor networks
Reference: Philip Lewis, ICSI Talk, May 2004
5
Towards a Sensor-Net
ArchitectureP
ower m
anagement
Mobility
managem
ent
Task
managem
ent
Application layer
Transport layer
Network layer
Data link layer
Physical layer
Traditional view of the
sensor network protocol
stack
(not strictly enforced)
Reference: Ian Akyildiz et al., Survey Paper, IEEE ComMag, Aug 2002
6
Towards a Sensor-Net
Architecture
● Alphabet soup of protocols and subsystems
● Widely differing assumptions about:
- the rest of the system and,
- how its part should interact
● Vertically integrated designs
- work with own set of components
- unable to inter-operate
● No standards that the protocols and solutions need to conform to
- good for research
- bad for interoperability
7
Sensor-Net (SP) Protocol
● One of the early encouraging steps towards a sensor-net architecture
● Unlike IP, SP sits between the network layer and the data link layer
REASON: processing potentially occurs at each hop not just at end points
● Allows multiple network protocols and link technologies to co-exist
● Abstraction could be implemented in any OS
● SP performs three main operations:
a) Data SEND
b) Data RECEIVE
c) Neighbor Management
● Main differences from IP
a) feedback e.g. Congestion, phase shift
b) network protocols can request urgent/reliable service
c) allow network and link layer to share link information
8
Towards a Sensor-Net
Architecture
Sensor-Net Protocol
Data Link
Physical Architecture
Tim
in
gSec
urit
yDis
cove
r
yPow
er
Man
agem
ent
Sys
tem
Man
agem
ent
Mob
ility
Man
agem
ent
Sensor-Net Application
In-Network Storage
Address-Free Protocols Name-Based Protocols
Sensing Carrier Sense
Media Access Time Stamping
Naming Graphs
Estimation
Transmit Receive
ACK
CachingSuppression
TriggersCustody Transfer
9
Sensor-Net (SP) Protocol
SP
Neighbor
Table
Msg
Pool
Neighbors Send Receive
SP Adaptor A SP Adaptor B
Data Link A Data Link B
Network
Protocol 1
Network
Protocol 2
Network
Protocol 3
Network
Service
Manager
10
SP vs ZigBee
Apart from SP there are other emerging standards as well e.g. ZigBee
“ZigBee proposes a classic layered architecture, but each layer assumes a
specific instance of the surrounding layers: e.g., the routing layer assumes
the IEEE 802.15.4 link and physical layers. An architecture build on static
technologies is destined for obsolescence”
Reference: Joe Polastre et al., “A Unifying Link Abstraction for Wireless Sensor Networks”, In
Proc. ACM SenSys 2005.
11
Mobility in Sensor Networks
● Research community generally ignores mobility in sensor networks
- they assume static sensor nodes
● Recent works have enabled mobility in sensor-nets
- e.g. RoboMote [Ref: K. Dantu et al., RoboMote paper, IPSN 2005],
- and Parasitic Mobility [Ref: MIT Media Lab, Parasitic Mobility paper, Pervasive 2005]
● Medical care or disaster response applications use mobile sensor nodes
- e.g. sensors attached to doctors or first responders
● Most protocols designed for static sensor networks perform poorly in mobile scenarios
- e.g. MAC protocols [Ref: M. Ali et al. MMAC paper, IEEE IPCCC 2005]
● Mobility could even improve other things like:
- coverage [Ref: B. Lie et al., Mobility Improves Coverage of Sensor Networks, Mobihoc 2005]
- localization [Ref: David Evans et al., Localization for Mobile Sensor Networks, Mobicom 2004]
12
Mobility in Sensor Networks
Image courtesy RobotMote – USC
Example of Hardware Mobility: A Group of RoboMotes
13
Mobility in Sensor Networks
Image courtesy CodeBlue - Harvard
Example of Medical Care and Disaster Relief Applications
14
Mobility-Management
● Mobility information could be required:
- at the application layer (e.g. monitoring physical movement of depression patients)
- at the network layer (e.g. neighbour discovery, route maintenance)
- at the MAC layer (e.g. MMAC: mobility adaptive MAC [IEEE IPCCC 2005])
● Protocols at different layers:
- could gather, store and manage mobility information individually (current practice)
- could make use of a cross-layer service that takes care of their mobility needs (our
proposal)
● Instead of exporting information between different layers (redundant) it is more useful to:
- import mobility information into a separate management database
- make this database visible across all layers
● Standardizing what goes into the database:
- enables network protocols and management applications to evolve independent of each other
- helps in moving towards a sensor-net architecture
16
Mobility-Management
● Our “bow-tie” mobility management design:
- does NOT take any stance on Time Synchronization (works with any)
- does NOT take any stance on naming (but assumes that nodes have unique addresses)
● Cross-layer database is implemented as a shared buffer and:
- is populated by information collected from the left-side of the bow-tie (SP network stack)
- provides services to management applications (right-side of the bow-tie)
● For mobility estimation:
- we propose to use AR-1 model [Ref: Z. Zaidi et al., Globecom 2004 and Secon 2004]
- more accurate AR-3 model is too computationally intensive for sensor nodes
● Accuracy of mobility estimation depends on underlying localization mechanism
● There is some communication overhead to gather and update mobility information of nodes
- is it worth it?
17
On-going Work
● Currently implementing SP and the mobility-management cross-layer service
- on Contiki Operating System [Ref: A Dunkels et al., Contiki paper, EmNets-I 2004]
- using Protothreads [Ref: A Dunkels et al., Protothreads paper, RealWSN 2005]
● For simulations:
- using COOJA simulator for Contiki [Ref: F. Osterlind, SICS Tech. Rep. T2006-05]
- using COOJA reduces the time to map simulation code to real deployments
● For mobility evaluations:
- implementing realistic mobility models [Ref: T. Camp et al., WCMC 2002]
- and using real mobility traces [Ref: D. Kotz et al., ACM MSWiM 2004]
18
Open Issues: Standard
Database?
Node ID Predicted (X,Y)
For Time Original
Time Stamp
7 (23,5) T1 + i T1
3 (102,17) T2 + j T2
15 (0,96) T3 + i T3
7 (24,6) T1 + j T1
19
Open Issues: IP over SP?
Sensor-Net Protocol
Data Link
Physical Architecture
Tim
in
gSec
urit
yDis
cove
r
yPow
er
Man
agem
ent
Sys
tem
Man
agem
ent
Mob
ility
Man
agem
ent
Sensor-Net Application
Sensing Carrier Sense
Media Access Time Stamping
Transmit Receive
ACK
Internet Protocol (IP)
TCP or UDP
Address-Free
Protocols
Name-Based
Protocols
20
Conclusions
● Current sensor-net literature
- presents an alphabet soup of protocols and sub-systems
- which do not inter-operate and make varying assumptions about others
● SP is an encouraging step towards a sensor network architecture
● Researchers assume “static sensor nodes” – an assumption that might not be valid now
● SP’s unifying link-abstraction and our mobility-management framework could:
- provide efficient mobility handling
- enable efforts from different research groups to inter-operate with each other
● Sensor-net community may make use of SP with mobility-management as a cross-layer service
to provide a standardized yet flexible framework for future research